How to switch between training and validation dataset with tf.MonitoredTrainingSession?

问题

I want to use feedable iterator design in tensorflow Dataset API, so I can switch to validation data after some training steps. But if I switched to validation data, it will end the whole session.

The following code demonstrate what I want to do:

import tensorflow as tf


graph = tf.Graph()
with graph.as_default():
    training_ds = tf.data.Dataset.range(32).batch(4)
    validation_ds = tf.data.Dataset.range(8).batch(4)

    handle = tf.placeholder(tf.string, shape=[])
    iterator = tf.data.Iterator.from_string_handle(
        handle, training_ds.output_types, training_ds.output_shapes)
    next_element = iterator.get_next()

    training_iterator = training_ds.make_initializable_iterator()
    validation_iterator = validation_ds.make_initializable_iterator()


with graph.as_default():

    with tf.train.MonitoredTrainingSession() as sess:
        training_handle = sess.run(training_iterator.string_handle())
        validation_handle = sess.run(validation_iterator.string_handle())
        sess.run(training_iterator.initializer)
        count_training = 0
        while not sess.should_stop():
            x = sess.run(next_element, feed_dict={handle: training_handle})
            count_training += 1
            print('{} [training] {}'.format(count_training, x.shape))
            # print(x)

            # we do periodic validation
            if count_training % 4 == 0:
                sess.run(validation_iterator.initializer)
                count_validation = 0
                while not sess.should_stop():
                    y = sess.run(next_element, feed_dict={handle: validation_handle})
                    count_validation += 1
                    print('  {} [validation] {}'.format(count_validation, y.shape))
                    # print(y)

The training data has 32 elements, batched with 4, so got 8 batches we do validation every 4 steps, so I expect:

#  1 [training]
# 2 [training]
# 3 [training]
# 4 [training]
#      1 [validation]
#      2 [validation]
# 5 [training]
# 6 [training]
# 7 [training]
# 8 [training]
#      1 [validation]
#      2 [validation]

but it stops when the first validation is done:

# 1 [training]
# 2 [training]
# 3 [training]
# 4 [training]
#      1 [validation]
#      2 [validation]

So, how to use this feedable iterator in tf.MonitoredTrainingSession?

回答1:

I would suggest to catch tf.errors.OutOfRangeError raised at the end of the validation dataset (you can also check the processing multiple epochs section in the official API for another solution using the repeat dataset ):

while not sess.should_stop():
    x = sess.run(next_element, feed_dict={handle: training_handle})
    count_training += 1
    print('{} [training] {}'.format(count_training, x.shape))

    # we do periodic validation
    if count_training % 4 == 0:
        sess.run(validation_iterator.initializer)
        count_validation = 0
        while True:
            try:
                y = sess.run(next_element, feed_dict={handle: validation_handle})
                count_validation += 1
                print('  {} [validation] {}'.format(count_validation, y.shape))
            except tf.errors.OutOfRangeError:
                break

This piece of code prints:

1 [training] (4,)  
2 [training] (4,)  
3 [training] (4,)  
4 [training] (4,)  
  1 [validation] (4,)  
  2 [validation] (4,)  
5 [training] (4,)
6 [training] (4,)
7 [training] (4,)
8 [training] (4,)
  1 [validation] (4,)
  2 [validation] (4,)

来源：https://stackoverflow.com/questions/49095849/how-to-switch-between-training-and-validation-dataset-with-tf-monitoredtrainings

标签

tensorflow

dataset

tensorflow-datasets

tensorflow-estimator